Cerebras CEO, Andrew Feldman on Why Raise $1BN and Delay the IPO & Why NVIDIA’s Worried About Growth

Cerebras CEO Andrew Feldman explains why he raised $1.1B instead of IPOing, why Nvidia’s tactics signal growth anxiety, and why chip depreciation is misunderstood.

• Cerebras raised $1.1B Series G at $8.1B valuation led by Fidelity; Fidelity’s participation sends a specific signal to Wall Street public-market confidence.
• Nvidia’s investments in OpenAI and predatory pre-announcements (B300 before B200 ships) are classic large-company tactics used when technical leadership is threatened.
• Real chip depreciation question is not age but whether next-gen performance/watt gain justifies retiring fully-paid-off hardware; H100s still in use beyond 2-year cycles.
• Memory bandwidth, not compute flops, is the fundamental inference bottleneck for GPU architectures; Cerebras solved this by building a wafer-scale chip filled with fast SRAM.
• UAE customers (G42 and others) accounted for ~75-80% of Cerebras revenue through H1 2024, placing orders so large they consumed all manufacturing capacity.
• TSMC fab construction takes years and costs $30-50B each; data center gigawatt facilities being announced are not yet operational, keeping chip supply below demand.
• AI talent shortage is a structural US bottleneck; immigration policy restricting H-1B visas directly harms the pipeline of engineers needed to meet compute demand.
• AlphaFold solved protein folding years ago yet no resulting drug exists yet — Feldman uses this to argue AI productivity diffusion takes 10-15 years, not 3-5.

2025-10-06 · Watch on YouTube